Update bert_layers.py

bert_layers.py

@@ -465,7 +465,7 @@ class BertEncoder(nn.Module):
                                                                 attn_mask=attention_mask,
                                                                 bias=alibi_attn_mask)
                 # JAANDOUI
-                print(f'Inner Attention: {attention_weights}')
+                # print(f'Inner Attention: {attention_weights}')
                 print(f'Inner Attention shape: {attention_weights.shape}')
                 all_attention_weights.append(attention_weights)  # Store attention weights
                 if output_all_encoded_layers:
@@ -501,8 +501,8 @@ class BertEncoder(nn.Module):
                                                               attn_mask=attention_mask,
                                                               bias=alibi_attn_mask)
             all_attention_weights.append(attention_weights)  # JAANDOUI: appending the attention of different layers together.
-            print(f'here is the matrix of attentions inside encoder: \n {all_attention_weights}')
-            print(f'and this is the shape inside encoder: \n {all_attention_weights.shape}')
+            # print(f'here is the matrix of attentions inside encoder: \n {all_attention_weights}')
+            print(f'and this is the shape inside encoder: \n {all_attention_weights[0].shape}')
             # print(f'NUMBER6: {all_attention_weights}')
         if not output_all_encoded_layers:
             all_encoder_layers.append(hidden_states)